This document outlines how to run tests for Panurus.
To work with Panurus and run tests, you first need to clone the repository and set up your environment.
Clone the code and make sure it is on your $GOPATH.
(Important: we assume in this documentation and default configuration that your $GOPATH has a single root-directory!).
Sometimes, we use $PANURUS_PATH to refer to Panurus repository in your filesystem.
export PANURUS_PATH=$GOPATH/src/github.com/LFDT-Panurus/panurus
git clone https://github.com/LFDT-Panurus/panurus.git $PANURUS_PATH
Before running tests, ensure you have the necessary tools and environment set up.
Panurus uses a system called NWO from Fabric Smart Client for its integration tests and samples to programmatically create a fabric network along with the fabric-smart-client nodes.
make install-tools
After installing the tools, run the checks to verify that everything is in order:
make checks
Download Fabric Binaries: The integration tests require Hyperledger Fabric binaries.
In order for a fabric network to be able to be created you need to ensure you have downloaded the appropriate version of the hyperledger fabric binaries from Fabric Releases and unpack the compressed file onto your file system. This will create a directory structure of /bin and /config. You will then need to set the environment variable FAB_BINS to the bin directory.
Do not store the fabric binaries within your panurus cloned repo as this will cause problems running the samples and integration tests as they will not be able to install chaincode.
Almost all the samples and integration tests require the fabric binaries to be downloaded and the environment variable FAB_BINS set to point to the directory where these binaries are stored. One way to ensure this is to execute the following in the root of the panurus project:
make download-fabric
export FAB_BINS=$PWD/../fabric/bin
You can also use this to download a different version of the fabric binaries, for example:
FABRIC_VERSION=2.5 make download-fabric
make testing-docker-images
make docker-images
You can run unit tests using the following make targets:
make unit-tests
make unit-tests-race
CI runs this suite as three parallel shards, which you can also run on their own:
make unit-tests-race-db (selector and storage packages; needs make testing-docker-images),
make unit-tests-race-core (token/core), and make unit-tests-race-rest (everything else).
make unit-tests-regression
Panurus uses Go’s native fuzzing (go test -fuzz) to exercise the stateless token
validators with adversarial byte input. Fuzz targets assert that validation entry points
never panic on malformed or truncated data, and — where applicable — that structurally
valid encodings are still correctly accepted.
| Target | Package | What it fuzzes |
|---|---|---|
FuzzVerifyTokenRequestFromRawNoPanic |
token/core/common |
Raw TokenRequest bytes fed straight into request verification. |
FuzzStructuredTokenRequestSignatureEnvelope |
token/core/common |
A structurally valid request with a mutated action-signature envelope (action ID, signature bytes, duplicate signatures). |
FuzzActionDeserializerNoPanic |
token/core/fabtoken/v1/validator |
A single issue or transfer action’s raw bytes through the FabToken ActionDeserializer. |
FuzzActionDeserializerNoPanic |
token/core/zkatdlog/nogh/v1/validator |
A single issue or transfer action’s raw bytes through the ZKAT-DLOG ActionDeserializer, including both Range and CSP proof-type encodings. |
FuzzActionDeserializerMultiActionNoPanic |
token/core/zkatdlog/nogh/v1/validator |
Two independently typed, independently fuzzed actions in the same TokenRequest. |
FuzzBulletProofVerifierNoPanic |
token/core/zkatdlog/nogh/v1/issue |
Raw BulletProof issue-proof bytes through BulletProofVerifier.Verify — the call IssueValidate makes on unauthenticated wire bytes. |
FuzzCSPVerifierNoPanic |
token/core/zkatdlog/nogh/v1/issue |
Raw CSP issue-proof bytes through CSPVerifier.Verify. |
FuzzBulletProofVerifierNoPanic |
token/core/zkatdlog/nogh/v1/transfer |
Raw BulletProof transfer-proof bytes through BulletProofVerifier.Verify — the call TransferValidate makes on unauthenticated wire bytes. |
FuzzCSPVerifierNoPanic |
token/core/zkatdlog/nogh/v1/transfer |
Raw CSP transfer-proof bytes through CSPVerifier.Verify. |
FuzzTokenDeserializeNoPanic |
token/core/zkatdlog/nogh/v1/token |
Raw ledger-stored output bytes through Token.Deserialize. |
FuzzMetadataDeserializeNoPanic |
token/core/zkatdlog/nogh/v1/token |
Raw ledger-stored metadata bytes through Metadata.Deserialize. |
FuzzPublicParamsDeserializeNoPanic |
token/core/zkatdlog/nogh/v1/setup |
Raw ledger-stored public parameters through NewPublicParamsFromBytes, exercised by every validator/prover/verifier on startup and params update. |
FuzzProofDeserializeNoPanic |
token/core/zkatdlog/nogh/v1/crypto/upgrade |
Raw untrusted token-upgrade request bytes through Proof.Deserialize. |
FuzzRegisterRecipientIdentityStructuralGuard |
token/services/identity/wallet |
A counterparty’s identity and audit-info bytes through Service.RegisterRecipientIdentity, asserting the structural guard rejects exactly the out-of-bounds lengths and never panics. |
Every fuzz target ships with an f.Add seed corpus of deterministic valid and malformed
inputs. These seeds run automatically as ordinary subtests whenever the package’s unit
tests run — no special flag is needed:
go test ./token/core/zkatdlog/nogh/v1/validator/...
# or, for the whole repo:
make unit-tests
To mutate the seed corpus and search for new crashing or panicking inputs, run a specific
target with -fuzz and bound the run with -fuzztime:
go test ./token/core/zkatdlog/nogh/v1/validator/... \
-fuzz=FuzzActionDeserializerNoPanic -fuzztime=30s
Omit -fuzztime to let the fuzzer run until you stop it (Ctrl+C). If it finds a
failing input, Go writes it to testdata/fuzz/<FuzzTestName>/<hash> in the target’s
package — commit that file to add the regression permanently to the seed corpus, then
re-run the failing case as a normal test with go test -run=<FuzzTestName>/<hash>.
.github/workflows/nightly-fuzz.yml runs every fuzz target above for up to 4h each
night (02:17 UTC) against main, as a fail-fast: false matrix so one crashing target
never stops the others. It can also be triggered manually via workflow_dispatch, which
accepts a fuzztime input to shorten or lengthen the campaign.
If a target finds a crashing or panicking input, the workflow:
testdata/fuzz/<FuzzTestName>/<hash>) and the full run log
as a build artifact, retained for 30 days;Nightly fuzz failure: <target> (labels testing,
bug), or comments on the existing one if that target already has an open issue, with
the run link and the local repro command.Confirmed crashes should have their seed file committed into the target’s
testdata/fuzz/<FuzzTestName>/ directory (see above) to add the regression permanently
to the corpus, closing out the issue once fixed.
maxFuzzActionBytes /
maxFuzzRequestBytes / maxFuzzSignatureBytes in each *_fuzz_test.go file) purely to
keep local runs fast. These caps are a fuzzing-harness convenience, not protocol rules.fuzzBlindingFactor in token/core/zkatdlog/nogh/v1/issue/verifier_fuzz_test.go,
via GetTokensWithWitnessAndBF), never the random GetTokensWithWitness. Some panics —
e.g. the cross-curve proof.u panic in the CSP range verifier — are only reached after a
proof’s proof-of-knowledge check passes, and that check binds to the token commitments.
With random per-run blinding factors a committed crash seed’s proof fails the PoK check
against freshly-randomised tokens and returns an error instead of panicking, so the seed
cannot reproduce the crash and the regression corpus is worthless. Any harness that feeds
a proof verifier must keep its tokens deterministic for committed seeds to be meaningful.Integration tests are crucial for verifying the interaction between different components.
Run specific integration tests using the integration-tests-<target> pattern.
Here are some common integration test targets (refer to .github/workflows/tests.yml or integration/ folder for a full list):
dlog-fabric-t1fabtoken-fabric-t1nft-dlognft-fabtokendvp-fabtokeninterop-fabtoken-t1To run the dlog-fabric-t1 test:
make integration-tests-dlog-fabric-t1
Most suites build a whole network for every spec with BeforeEach(ts.Setup) /
AfterEach(ts.TearDown). Specs that leave the network healthy can instead share one network
by calling token.ShareNetwork (integration/token/test_utils.go) inside an
Ordered, ContinueOnFailure container, the Panurus counterpart of FSC’s
faultrecovery.ShareNetwork:
Describe("Multisig and Redeem to yourself", Ordered, ContinueOnFailure, t.Label, func() {
ts, selector := newTestSuite(...)
token2.ShareNetwork(ts)
It("Multisig succeeded", Label("T12"), func() { fungible.TestMultiSig(ts.II, selector) })
It("Test redeem", Label("T13"), func() { fungible.TestRedeem(ts.II, selector, "default") })
})
The lifecycle is:
BeforeAll, before the first selected spec. The
DeferCleanup that tears it down is registered before the build starts, so a network whose
build failed part-way is still torn down and does not hold the ports that later containers of
the suite reuse. If the initial build fails, Ginkgo fails that spec and skips the rest of the
container.DeferCleanup it registered
itself, may leave the network broken, so the network is torn down and rebuilt before the next
spec runs. If that rebuild fails, the spec it was run for fails before its body runs (it is
not skipped, so the run stays red), and every later spec of the container is skipped rather
than retrying the rebuild.--fail-fast, Ginkgo skips every spec after the first failure and
never runs that final cleanup, so the network is torn down right after the spec that failed.faultrecovery.ShareNetwork. A CI runner
is discarded anyway; locally, run make clean-all-containers.Because of this, the suite passed to token.ShareNetwork must implement token.Network
with a TearDown that is safe to call on a network whose Setup failed part-way (or never
ran) and that is idempotent. token.TestSuite does so; a suite embedding FSC’s
*integration.TestSuite must override TearDown, for example with token.StopInfrastructure
(see the testSuite of integration/token/fungible/dlogx). TearDown clears II only once
Stop returns, so if Stop panics part-way the final cleanup retries it once. It does not stop
the build server or remove the test directory of an infrastructure whose Generate never ran;
those hold no ports in the start range and are cleaned up by CleanupBuildArtifacts and
DeleteOnStart.
Labels stay on the specs, so a label filter that selects one spec still runs it on a network of
its own; a CI job gains only when its filter selects several specs of the container
(e.g. make integration-tests-dlog-fabric-t12-t13).
Only group specs that:
TestAllfungible.TestAll asserts the whole audited and accepted transaction history, the issuer’s
history and the issuer’s USD quota, so it only passes on a fresh network. The dlog suite
(integration/token/fungible/dlog) still shares its network with the cheaper specs of the same
topology, in the containers “T1 Fungible with Auditor ne Issuer and extras” (T1, T17, T4, T5,
T3, T2.1, T2) and “Fungible with Auditor = Issuer” (T6, T7); the fabtoken suite does the same in
its “Fungible” container (T1, T17, T4, T5, T3, T2). The rules are:
TestAll runs first. It checks this itself: it fails at once, with “TestAll must run on a
fresh network (first spec of a ShareNetwork container)”, if the auditor has already audited a
transaction. The check only looks at that auditor’s audit history; its point is a clear error,
since TestAll’s own first history assertions would fail anyway.fungible.WithTokenType, accepted by
TestRedeemUnauthorizedIssuer, TestRemoteOwnerWallet(WithWMP), TestRevokeIdentity and
TestPublicParamsUpdate (the default stays "USD"; TestPublicParamsUpdate also takes
fungible.WithRestartAfterUpdate). Balances, holdings, spending
and the auditor’s limits are per type, and the USD quota applies to the exact type "USD"
only, so a spec with its own type does not see what TestAll left behind. Two specs that
assert absolute balances of the same wallets (T4 and T5) need different types from each other
too. The type must not contain .: the sherdlock selector keys its cache by wallet ID and type
joined by ., and wallet IDs are dotted (alice.id1). The suites use USDt17, USDt4, and
so on.TestAll already did; that is a KVS overwrite, so it changes nothing. A spec that
needs the absence of such state (T13 expects no authorized redeem issuer) cannot run after
TestAll. Likewise T17 relies on newIssuer not being in the public parameters, so it runs
before T2.1/T2, which add it.PrepareUpdatedPublicParams reads
the public parameters file written at network generation, which UpdatePublicParams does not
rewrite, so T2.1 and T7 drop the issuer.ExtraId issuer that TestAll added. No later spec
uses it; a spec that needs an earlier update to persist must not follow one.To pick the order for a new container: start with TestAll, then the specs that only add state
under their own token type, then those that change shared state (revocations), then those that
change the public parameters, and check each against the ones before it with the rules above.
The make targets integration-tests-dlog-fabric-aries-{websocket,libp2p,replicas},
integration-tests-dlog-fabric-t6-t7-{websocket,libp2p,replicas} and
integration-tests-fabtoken-fabric-all-{websocket,libp2p,replicas} run each container once per
infrastructure type; the single-label targets (e.g. integration-tests-dlog-fabric-t17) still
run one spec on a network of its own.
Co-locating TestAll with the small specs is a tradeoff: it cuts the dlog jobs for these specs
from 9 to 3, but each grouped job runs TestAll plus every small spec back to back, so the
aries-replicas job becomes the longest of them (estimated 22–28 minutes, well under Ginkgo’s
1h default suite timeout). If it becomes the critical path of the itest workflow, split the
container into TestAll alone plus a shared container for the small specs.
The Fabric-X suite (integration/token/fungible/dlogx) shares networks too, with two
differences from the dlog suite:
TestAll-first container for the small specs. Its “Extras” container (T12, T13, T4,
T5, T3) has no TestAll in front: TestAll authorizes the issuer for redeem, which T13
needs not to have happened, and the fabricx TestAll jobs (T1, T6, T16) are already the
longest of the itest workflow, so adding specs behind them would lengthen the critical path.
The container follows the rules above: T12 and T13 issue 220 USD between them, within the
USD quota; T4, T5 and T3 each use their own token type; T3 runs after T4 and T5, and last.
“Extras” runs on the Aries|WithEndorsers|WebEnabled topology, so T12 and T13 now also get
the FSC web server, which they do not use.issuer node holds
(signPublicParamsWithCurrentIssuer in
token/services/network/fabric/endorsement/fsc/service.go). T2 and T7 replace the issuers
with newIssuer, so nothing can update the public parameters after them. T2.1 appends
newIssuer, so T2 should be able to follow it, but that order has not been run on Fabric-X.T2.1, T2 and T7 failed locally on Fabric-X, each on its own network (a GetPublicParams
timeout right after the update; see
#1945). T2.1 and T2 therefore stay in a
container with a network per spec (“Public params update”), and T6 and T7 do not share a
network; all three are disabled in CI anyway. Once they pass, add T2.1 then T2 after T3 in
“Extras”, each with its own token type, and T7 after T6.
CI runs the specs of “Extras” it enabled before, T13, T4, T5 and T3, with
make integration-tests-fabricx-dlog-extras: all three infrastructure types in one job, one
network each; -extras-{websocket,libp2p,replicas} run one type each. To enable T12, T2.1 or
T2 in CI, widen FABRICX_DLOG_EXTRAS_FILTER in fabricx.mk rather than adding a job, and
split the job per infrastructure type if it then gets close to the longest itest jobs.
By default, an FSC node only gets token: enabled: true written into its generated
core.yaml if it belongs to at least one TMS (Topology.AddTMS). Topology.EnableTokenPlatform
(integration/nwo/token/topology.go) lets a topology install and start the token platform on
every node of an fsc.Topology with no TMS defined at all:
tokenTopology := token.NewTopology()
tokenTopology.EnableTokenPlatform(fscTopology)
This only affects configuration generation — the base token: stanza from the extension
template, without any TMS-specific selector or driver configuration. It does not attach the
token Go SDK to the nodes; do that separately with Topology.SetSDK or
fscTopology.AddSDK(...), as usual.
The token SDK starts up cleanly with zero TMS configs (all its dependencies are lazily
provided and network.Provider.Connect() is a no-op with no configured networks), but
GetManagementService(...) still returns “no token management service configs found” until a
TMS is registered. EnableTokenPlatform is meant as the config-generation groundwork for
bringing up nodes that will have a TMS registered dynamically at runtime, once that
capability exists — it is not itself a way to use tokens.
For tests involving Fabric-X (starts with fabricx), you need additional setup:
make fxconfig configtxgen fabricx-docker-images
make integration-tests-fabricx-dlog-t1
After running tests, especially integration tests that spin up Docker containers, you might want to clean up your environment.
# Clean up Docker artifacts (containers, volumes, networks) and generated files
make clean
# Remove all Docker containers (running and stopped)
make clean-all-containers
# Remove Fabric peer images
make clean-fabric-peer-images